contador Saltar al contenido

Intel has trained a neuromorphic chip to recognize ten different types of odors

Intel and Cornell University have published a joint study demonstrating some interesting capabilities of "Loih", A neuromorphic chip capable of learning and recognizing various dangerous substances by smell, even in the presence of" significant "abstract noise of data and occlusions. The study co-authors report that the neuromorphic computing it could be used to detect precursor odors of explosives, detect narcotics, recognize polymers and other areas.

In the study, disclosed in the scientific publication Nature Machine Intelligence, Intel and Cornell university researchers, describe the possibility of "teaching" Loihi to recognize various odors by configuring the olfactory circuit diagram by drawing on a data set that allows monitoring the response of 72 different chemical sensors to various types of miasms. The researchers further explain that their technique offers a higher level of recognition than conventional cutting-edge methods, highlighting – among other things – a machine learning mechanism that required 3,000 times greater training activities to obtain high levels of accuracy in classification.

Intel has trained a neuromorphic chip to recognize ten different types of odors

Intel's Loihi chip built with 14nm technology integrates a die 60mm, over 2 billion transistors, 130,000 artificial neurons, 130 million synapses and three Lakemont cores for "orchestration". Peculiarity of Loihi is the programmable "engine" for on-chip training of neural networks spiking (SSN) used to actually mimic natural neural networks.

According to Intel (more details here), Loihi processes information up to 1000 times faster, 10,000 times more efficiently than traditional processors, and can solve some optimization problems with gains of up to three orders of magnitude in terms of speed and efficiency energy.

The growing number of data produced imposes the need for an infrastructure capable of managing them. The need to collect and analyze huge amounts of data and make decisions in real time is directing various companies (including Google and IBM) towards the development of new and innovative forms of computing that are more efficient than the current CPU architecture. One of these innovations is neuromorphic computing, which involves applying the principles of the biological brain to computer architectures to create self-learning systems.

Learning ability allows neuromorphic chips to perform more computing at a much faster rate than possible with current chips. These next generation processors will be important for the future of artificial intelligence and could be used for data processing in changing real-time environments. Although the neuromorphic computing market is still in its infancy, analysts predict that it could go from $ 69 million in 2024 to $ 5 billion in 2029, reaching a staggering $ 21.3 billion by 2034.

For all the articles of macitynet that speak of Artificial Intelligence refer to this address.